Nvidia working on CUDA-based anti-virus | Hacker News Reader

Nvidia working on CUDA-based anti-virus

bit-tech.net

11 pointschanux10 comments

What an awkward thing to off-load to the GPU. Isn't virus scanning bound by I/O, and trailing far behind by substring matching?

Maybe there are some very processor-intensive heuristics that are not being used because they don't want to impact CPU performance.

I can't seeing this be effective in any other situation. My understanding is that the bandwidth to and from the GPU is the big bottleneck. Once its on the card, the processing is fast, but getting things into the GPU memory is the bottleneck.

If, on the other hand, they have some sort of complex heuristics that is heavily process intensive, this may be very successful. As it stands, I don't think (although IANASecurity Expert) a definition based solution will work do to the memory limitations.

I don't think the transfer to GPU memory is so much a bottleneck as it involves some latency. The PCIe bus provides a ton of bandwidth, it just takes longer to get stuff there from main memory than it does to get stuff to the CPU from main memory.

So, wasn't trying to pick nits, just trying to say that if it's a latency issue, the penalty is roughly the same for small amounts of data as it is for big amounts, and if you have a huge amount of data that takes non-trivial amount of processing, the latency penalty is, as you said, worth it.

If I recall correctly, AGP, the older graphics interface was very lopsided with respect to bandwidth (very fast to send data to the card, very slow to get data back). As far as I understand, the newer PCI Express (newer == several years old now), has more balanced (and fast) bandwidth.

trivially parallelizable processor-intensive heuristics?

While true, this means that the CPU can be used for other things.

Also I doubt its I/O bound because there is still lots of scanning to do (but I am not 100% sure). Also with fast hard drives (or SSDs) it might be CPU bound.

Also, when running applications currently the CPU must load the app, scan it, then run it, while the GPU can probably do it faster thus making the system overall more responsive.

Wait... Also, the real-time scanner is limited. It does not do a full check on every file because that is too expensive (though it might be damn close), with the GPU they might have the power to make scanning through your whole computer 99% moot since every time any program executes the virus scanner gives it a 100% scan, fast, without the user even noticing that it happened.

I can see lots of benefits of this. Go nvidia.

I am hoping nvidia kicks intel to the stone age with a new processor and such. Or at least comes up with some REAL competition, something that AMD started but could not maintain momentum. This is probably step 1. :)

Just a few points:

  - relative to clock speed, SSDs are still a magnitude slower.
  - Loading the app is a 100% i/o operation. "scan it, then run it": your gpu 
    will not run your app. The scanning part, fine let's assume the GPU can do 
    it. You still have to DMA it over to the GPU which involves a lot. Still 
    not convinced it'll be faster than the cpu.
  - "without the user ever noticing that it happened".. you still have to copy
    the text pages over after being loaded, good luck not getting the user to 
    notice it.

Anywho, I am not an AV expert nor do I completely understand what they're trying to do with GPUs, these are just some thoughts I have off the top of my head.

What I like about this is offloading the job to a processor that's generally not doing much. If you want to start taxing the gpu (playing a game, editing video, etc) you'll want anti-virus to stop running (or move into the background) at any rate.

Wow, now those Micrsoft ads for "cheaper laptops" can add a line that says, "Now you get an anti-virus processor included for FREE!"

Reply on news.ycombinator.com